The Threshold Problem of GMM

[joeyslv]

After carefully reading the article
3.4 Temporary consistency using Gaussian Mixture Model (GMM)
You use a Gaussian mixture distribution for each category. Each mixture distribution contains a positive and negative Gaussian distribution. It is also a posterior probability probability of EM algorithm inference for each category. However, in model training, the Log log only has a single GMM threshold, and ultimately GMM generates a category threshold or a unified threshold
Looking forward to your answer
Thank you.

[Johnson-Wang]

Hi,
Thanks for your interest. Only the mean gmm_thr value for all categories is logged to be printed out. Bounding boxes are filtered with a per-class threshold during training. Please refer to code https://github.com/Adamdad/ConsistentTeacher/blob/b49a9ce1c0ae78d13528be0a9a4eff65a627d581/ssod/models/consistent_teacher.py#LL335C25-L335C25 for details.

[joeyslv]

Hi, Thanks for your interest. Only the mean gmm_thr value for all categories is logged to be printed out. Bounding boxes are filtered with a per-class threshold during training. Please refer to code https://github.com/Adamdad/ConsistentTeacher/blob/b49a9ce1c0ae78d13528be0a9a4eff65a627d581/ssod/models/consistent_teacher.py#LL335C25-L335C25 for details.

Thank you for your answer. Let me take a closer look at this code